Graph-based Methods Research Articles

A hypergraph transformer method for brain disease diagnosis

ObjectiveTo address the high-order correlation modeling and fusion challenges between functional and structural brain networks.MethodThis paper proposes a hypergraph transformer method for modeling high-order correlations between functional and structural brain networks. By utilizing hypergraphs, we can effectively capture the high-order correlations within brain networks. The Transformer model provides robust feature extraction and integration capabilities that are capable of handling complex multimodal brain imaging.ResultsThe proposed method is evaluated on the ABIDE and ADNI datasets. It outperforms all the comparison methods, including traditional and graph-based methods, in diagnosing different types of brain diseases. The experimental results demonstrate its potential and application prospects in clinical practice.ConclusionThe proposed method provides new tools and insights for brain disease diagnosis, improving accuracy and aiding in understanding complex brain network relationships, thus laying a foundation for future brain science research.

Almost exact recovery in noisy semi-supervised learning

Abstract Graph-based semi-supervised learning methods combine the graph structure and labeled data to classify unlabeled data. In this work, we study the effect of a noisy oracle on classification. In particular, we derive the maximum a posteriori (MAP) estimator for clustering a degree corrected stochastic block model when a noisy oracle reveals a fraction of the labels. We then propose an algorithm derived from a continuous relaxation of the MAP, and we establish its consistency. Numerical experiments show that our approach achieves promising performance on synthetic and real data sets, even in the case of very noisy labeled data.

Graph-Based Methods for Forecasting Realized Covariances

Abstract We forecast the realized covariance matrix of asset returns in the U.S. equity market by exploiting the predictive information of graphs in volatility and correlation. Specifically, we augment the Heterogeneous Autoregressive model via neighborhood aggregation on these graphs. Our proposed method allows for the modeling of interdependence in volatility (also known as spillover effect) and correlation, while maintaining parsimony and interpretability. We explore various graph construction methods, including sector membership and graphical LASSO (for modeling volatility), and line graph (for modeling correlation). The results generally suggest that the augmented model incorporating graph information yields both statistically and economically significant improvements for out-of-sample performance over the traditional models. Such improvements remain significant over horizons up to 1 month ahead, but decay in time. The robustness tests demonstrate that the forecast improvements are obtained consistently over the different out-of-sample sub-periods and are insensitive to measurement errors of volatilities.

Boosting point cloud understanding through graph convolutional network with scale measurement and high-frequency enhancement

Graph-based methods have exhibited exceptional performance in point cloud understanding by capturing local geometric relationships. However, existing approaches often struggle to characterize the overall spatial scale of local graphs. In addition, they fail to capture the differences between nodes effectively, which is crucial for distinguishing different classes. This study introduces SM-HFEGCN, a novel graph convolutional network that addresses these limitations through two key innovations: scale measurement and high-frequency enhancement. First, we introduce a spatial scale feature derived from the diagonal vectors of the neighborhood, which serves as a unique graph-specific property related to the geometry and density of the local point cloud. This feature can characterize the overall spatial scale of the local point cloud. Second, we enhance the high-frequency information to capture node variations and integrate it with smoothed information to represent the differences and similarities between nodes simultaneously. Extensive experiments demonstrate the effectiveness of SM-HFEGCN in point cloud classification and segmentation tasks.

A Structured Bipartite Graph Learning method for ensemble clustering

Given a set of base clustering results, conventional bipartite graph-based ensemble clustering methods typically require computing a sample-cluster similarity matrix from each base clustering result. These matrices are then either concatenated or averaged to form a bipartite weight matrix, which is used to create a bipartite graph. Graph-based partition techniques are subsequently applied to this graph to obtain the final clustering result. However, these methods often suffer from unreliable base clustering results, making it challenging to identify a clear cluster structure due to the variations in cluster structures across the base results. In this paper, we propose a novel Structured Bipartite Graph Learning (SBGL) method. Our approach begins by computing a sample-cluster similarity matrix from each base clustering result and constructing a base bipartite graph from each of these matrices. We assume these base bipartite graphs contain a set of latent clusters and project them into a set of sample-latent-cluster bipartite graphs. These new graphs are then ensembled into a bipartite graph with a distinct cluster structure, from which the final set of clusters is derived. Our method allows for different numbers of clusters across base clusterings, leading to improved performance. Experimental results on both synthetic and real-world datasets demonstrate the superior performance of our new method.

Adaptive scheduling for Internet of Vehicles using deconfounded graph transfer learning

With the rapid development of Internet of Things technology, the Internet of Vehicles has emerged as a significant application. The Internet of Vehicles connects vehicles with traffic infrastructure and other vehicles, enabling real-time information sharing and vehicle dispatching. However, in the Internet of Vehicles environment, traffic scheduling systems face challenges such as sparse raw data and environmental confounding bias, which limit the effectiveness of traditional graph-based reinforcement learning methods. In this paper, we propose a Cross-City Federated Continual Learning Framework for Spatiotemporal Graph Transfer Learning called CCFTL, which removes confounding effects on reinforcement learning and enhances few-shot transfer learning in data-scarce cities. Our approach aims to significantly improve the adaptability and scalability of cross-city vehicle networking technologies in ever-changing environments. This study acts as an innovative research of the deconfounding strategy with spatiotemporal graph meta-knowledge learning, which enables optimization of cross-city meta-knowledge transfer through a federated continual learning approach. This approach effectively reduces learning bias caused by regional differences, which enhances the model’s generalization and adaptability to complex, heterogeneous urban scenarios. Experimental results show that our framework significantly outperforms traditional methods, especially in improving the efficiency and accuracy of traffic dispatch systems in Internet of Vehicles environments with scarce data and the presence of confounding factors.

Ensemble-Enhanced Semi-Supervised Learning with Optimized Graph Construction for High-Dimensional Data.

Graph-based methods have demonstrated exceptional performance in semi-supervised classification. However, existing graph-based methods typically construct either a predefined graph in the original space or an adaptive graph within the output space, which often limits their ability to fully utilize prior information and capture the optimal intrinsic data distribution, particularly in high-dimensional data with abundant redundant and noisy features. This paper introduces a novel approach: Semi-Supervised Classification with Optimized Graph Construction (SSC-OGC). SSC-OGC leverages both predefined and adaptive graphs to explore intrinsic data distribution and effectively employ prior information. Additionally, a graph constraint regularization term (GCR) and a collaborative constraint regularization term (CCR) are incorporated to further enhance the quality of the adaptive graph structure and the learned subspace, respectively. To eliminate the negative effect of constructing a predefined graph in the original data space, we further propose a Hybrid Subspace Ensemble-enhanced framework based on the proposed Optimized Graph Construction method (HSE-OGC). Specifically, we construct multiple hybrid subspaces, which consist of meticulously chosen features from the original data to achieve high-quality and diverse space representations. Then, HSE-OGC constructs multiple predefined graphs within hybrid subspaces and trains multiple SSC-OGC classifiers to complement each other, significantly improving the overall performance. Experimental results conducted on various high-dimensional datasets demonstrate that HSE-OGC exhibits outstanding performance.

Integrating higher-order relations for enhanced twitter bot detection

AbstractBot detection is a critical task in preserving the integrity of social networks and mitigating online disinformation. Despite advances in graph-based methods for detecting bots, these models often rely on follow relations, assuming that users sharing similar characteristics are more likely to connect. However, these methods often fail to address complex behavioral patterns indicative of coordinated bot activities, allowing bot developers to avoid detection more easily. We investigate the potential of integrating behavioral and higher-order relations, focusing on Retweet, Co-Retweet (where two users retweet the same tweet) and Co-Hashtag (where two users frequently use the same hashtag), and compare these against conventional follower and following relations. We further explore the process of relation creation, particularly highlighting the Co-Hashtag relation’s robustness against data collection flaws and its ability to mitigate shortcomings in the collection of datasets. Our experiments not only contribute to the ongoing efforts to address challenges in bot detection, but also open up new avenues for exploring how complex network patterns can be employed for social network analysis.

StratoMod: predicting sequencing and variant calling errors with interpretable machine learning

Despite the variety in sequencing platforms, mappers, and variant callers, no single pipeline is optimal across the entire human genome. Therefore, developers, clinicians, and researchers need to make tradeoffs when designing pipelines for their application. Currently, assessing such tradeoffs relies on intuition about how a certain pipeline will perform in a given genomic context. We present StratoMod, which addresses this problem using an interpretable machine-learning classifier to predict germline variant calling errors in a data-driven manner. We show StratoMod can precisely predict recall using Hifi or Illumina and leverage StratoMod’s interpretability to measure contributions from difficult-to-map and homopolymer regions for each respective outcome. Furthermore, we use Statomod to assess the effect of mismapping on predicted recall using linear vs. graph-based references, and identify the hard-to-map regions where graph-based methods excelled and by how much. For these we utilize our draft benchmark based on the Q100 HG002 assembly, which contains previously-inaccessible difficult regions. Furthermore, StratoMod presents a new method of predicting clinically relevant variants likely to be missed, which is an improvement over current pipelines which only filter variants likely to be false. We anticipate this being useful for performing precise risk-reward analyses when designing variant calling pipelines.

Pore size classification and prediction based on distribution of reservoir fluid volumes utilizing well logs and deep learning algorithm in a complex lithology

Pore size analysis plays a pivotal role in unraveling reservoir behavior and its intricate relationship with confined fluids. Traditional methods for predicting pore size distribution (PSD), relying on drilling cores or thin sections, face limitations associated with depth specificity. In this study, we introduce an innovative framework that leverages nuclear magnetic resonance (NMR) log data, encompassing clay-bound water (CBW), bound volume irreducible (BVI), and free fluid volume (FFV), to determine three PSDs (micropores, mesopores, and macropores). Moreover, we establish a robust pore size classification (PSC) system utilizing ternary plots, derived from the PSDs.Within the three studied wells, NMR log data is exclusive to one well (well-A), while conventional well logs are accessible for all three wells (well-A, well-B, and well-C). This distinction enables PSD predictions for the remaining two wells (B and C). To prognosticate NMR outputs (CBW, BVI, FFV) for these wells, a two-step deep learning (DL) algorithm is implemented. Initially, three feature selection algorithms (f-classif, f-regression, and mutual-info-regression) identify the conventional well logs most correlated to NMR outputs in well-A. The three feature selection algorithms utilize statistical computations. These algorithms are utilized to systematically identify and optimize pertinent input features, thereby augmenting model interpretability and predictive efficacy within intricate data-driven endeavors. So, all three feature selection algorithms introduced the number of 4 logs as the most optimal number of inputs to the DL algorithm with different combinations of logs for each of the three desired outputs. Subsequently, the CUDA Deep Neural Network Long Short-Term Memory algorithm(CUDNNLSTM), belonging to the category of DL algorithms and harnessing the computational power of GPUs, is employed for the prediction of CBW, BVI, and FFV logs. This prediction leverages the optimal logs identified in the preceding step. Estimation of NMR outputs was done first in well-A (80% of data as training and 20% as testing). The correlation coefficient (CC) between the actual and estimated data for the three outputs CBW, BVI and FFV are 95%, 94%, and 97%, respectively, as well as root mean square error (RMSE) was obtained 0.0081, 0.098, and 0.0089, respectively. To assess the effectiveness of the proposed algorithm, we compared it with two traditional methods for log estimation: multiple regression and multi-resolution graph-based clustering methods. The results demonstrate the superior accuracy of our algorithm in comparison to these conventional approaches. This DL-driven approach facilitates PSD prediction grounded in fluid saturation for wells B and C.Ternary plots are then employed for PSCs. Seven distinct PSCs within well-A employing actual NMR logs (CBW, BVI, FFV), in conjunction with an equivalent count within wells B and C utilizing three predicted logs, are harmoniously categorized leading to the identification of seven distinct pore size classification facies (PSCF). this research introduces an advanced approach to pore size classification and prediction, fusing NMR logs with deep learning techniques and extending their application to nearby wells without NMR log. The resulting PSCFs offer valuable insights into generating precise and detailed reservoir 3D models.

Deep Learning and Recurrent Signature Based Classification for Sensor-Based HAR: Addressing Explainability and Complexity in 5G Networks

When it comes to clinical applications, sensor-based human activity recognition (HAR) is invaluable, and numerous machine learning algorithms have effectively used to obtain excellent presentation. Using a variety of on-body sensors, these systems attempt to ascertain the subject's status relative to their immediate surroundings. There was a time when feature extraction was done by hand, but now more and more people are using Artificial Neural Networks (ANNs). A number of innovative approaches to HAR have surfaced since the advent of deep learning. Problems arise, however, for sensor-based HAR classification algorithms in today's communication networks. Among these, you can find solutions to problems like deal with complicated and large-scale data signals, extract characteristics from complicated datasets, and meet explainability standards. For complicated 5G networks, these difficulties become even more apparent. In particular, explainability is now critical for the broad use of sensor-based HAR in 5G networks and beyond. The research suggests a classification approach based on path signatures, recurrent signature (ReS), to address these issues. This cutting-edge model employs deep-learning (DL) approaches to circumvent the tedious feature selection challenge. Furthermore, the study investigates how to improve the ReS model's classification accuracy by using graph-based optimisation methods. To test how well the suggested framework worked, to dug deep into the publicly available dataset, which included a separate set of tasks. The paper's empirical results on AReM datasets achieved an average accuracy of 96%.

Dual-stream dynamic graph structure network for document-level relation extraction

Extracting structured information from unstructured text is crucial for knowledge management and utilization, which is the goal of document-level relation extraction. Existing graph-based methods face issues with information confusion and integration, limiting the reasoning capabilities of the model. To tackle this problem, a dual-stream dynamic graph structural network is proposed to model documents from various perspectives. Leveraging the richness of document information, a static document heterogeneous graph is constructed. A dynamic heterogeneous document graph is then induced based on this foundation to facilitate global information aggregation for entity representation learning. Additionally, the static document graph is decomposed into multi-level static semantic graphs, and multi-layer dynamic semantic graphs are further induced, explicitly segregating information from different levels. Information from different streams is effectively integrated via an information integrator. To mitigate the interference of noise during the reasoning process, a noise regularization mechanism is also designed. The experimental results on three extensively utilized publicly accessible datasets for document-level relation extraction demonstrate that our model achieves F1 scores of 62.56%, 71.1%, and 86.9% on the DocRED, CDR, and GDA datasets, respectively, significantly outperforming the baselines. Further analysis also demonstrates the effectiveness of the model in multi-entity scenarios.

Toward Explainable Affective Computing: A Review.

Affective computing has an unprecedented potential to change the way humans interact with technology. While the last decades have witnessed vast progress in the field, multimodal affective computing systems are generally black box by design. As affective systems start to be deployed in real-world scenarios, such as education or healthcare, a shift of focus toward improved transparency and interpretability is needed. In this context, how do we explain the output of affective computing models? and how to do so without limiting predictive performance? In this article, we review affective computing work from an explainable AI (XAI) perspective, collecting and synthesizing relevant papers into three major XAI approaches: premodel (applied before training), in-model (applied during training), and postmodel (applied after training). We present and discuss the most fundamental challenges in the field, namely, how to relate explanations back to multimodal and time-dependent data, how to integrate context and inductive biases into explanations using mechanisms such as attention, generative modeling, or graph-based methods, and how to capture intramodal and cross-modal interactions in post hoc explanations. While explainable affective computing is still nascent, existing methods are promising, contributing not only toward improved transparency but, in many cases, surpassing state-of-the-art results. Based on these findings, we explore directions for future research and discuss the importance of data-driven XAI and explanation goals, and explainee needs definition, as well as causability or the extent to which a given method leads to human understanding.

INCOMPLETE multi-view clustering based on low-rank adaptive graph learning

The challenge of acquiring complete data has led to substantial progress in incomplete multi-view clustering (IMVC) methods. Because graph structures can be excellent representations of data structure relationships, exceptional performance in handling incomplete data is demonstrated by graph-based methods at present. However, these methods still have their limitations. Most incomplete multi-view algorithms primarily focus on local information, neglecting global information. Therefore, these methods cannot dynamically recover the structural relationships in incomplete data by harnessing potential information from multiple perspectives and overall structural information. In response to the aforementioned concerns, we introduced an IMVC based on low-rank adaptive graph learning (IMVC-LAGL). This method initially constructs an affinity matrix based on the inter-view adjacency relationships. It also utilizes tensor low-rank constraints and consensus representation learning to explore higher-order correlations among different views. Subsequently, it adaptively reconstructs the incomplete graph structure to ultimately obtain a complete affinity relationship. It leads to excellent clustering results by integrating relevant information within views, overall structural information and potential information from multiple perspectives. We conducted experiments comparing our algorithm with eight incomplete multi-view algorithms using five different evaluation metrics. The results show that our algorithm achieves the best clustering results across eight datasets with varying missing rates. Particularly in the BBCSport dataset and YaleB dataset, the clustering accuracy of our algorithm is improved by 19.83 % and 16.41 %, respectively, compared with the second-best algorithm, under a 50 % missing rate.

Locally supported, quasi-interpolatory bases for the approximation of functions on graphs

Graph-based approximation methods are of growing interest in many areas, including transportation, biological and chemical networks, financial models, image processing, network flows, and more. In these applications, often a basis for the approximation space is not available analytically and must be computed. We propose perturbations of Lagrange bases on graphs, where the Lagrange functions come from a class of functions analogous to classical splines. The basis functions we consider have local support, with each basis function obtained by solving a small energy minimization problem related to a differential operator on the graph. We present ℓ∞ error estimates between the local basis and the corresponding interpolatory Lagrange basis functions in cases where the underlying graph satisfies an assumption on the connections of vertices where the function is not known, and the theoretical bounds are examined further in numerical experiments. Included in our analysis is a mixed-norm inequality for positive definite matrices that is tighter than the general estimate ‖A‖∞≤n‖A‖2.

Graph contrast learning for recommendation based on relational graph convolutional neural network

Current knowledge graph-based recommendation methods heavily rely on high-quality knowledge graphs, often falling short in effectively addressing issues such as the cold start problem and heterogeneous noise in user interactions. This leads to biases in user interest and popularity. To overcome these challenges, this paper introduces a novel recommendation approach termed Knowledge-enhanced Perceptive Graph Attention with Graph Contrastive Learning (KPA-GCL), which leverages relational graph convolutional neural networks. The proposed method optimizes the triplet embedding representation of entity-item interactions based on relationships between adjacent entities in a heterogeneous graph. Subsequently, a graph convolutional neural network is employed for enhanced aggregation. Similarity scores from a contrastive view serve as the selection criterion for high-quality embedded representations, facilitating the extraction of refined knowledge subgraphs. Multiple adaptive contrast-loss optimization functions are introduced by combining Bayesian Personalized Ranking (BPR) and hard negative sampling techniques. Comparative experiments are conducted with ten popular existing methods using real public datasets. Results indicate that the KPA-GCL method outperforms compared methods in all datasets based on Recall, NDCG, Precision, and Hit-ratio measures. Furthermore, in terms of mitigating cold start and noise, the KPA-GCL method surpasses other ten methods. This validates the reasonability and effectiveness of KPA-GCL in real-world datasets.

Graph-based insider threat detection: A survey

Insider threat detection has been a significant topic in recent years. However, as network technology develops, the intranet becomes more complex. Therefore, simply matching attack patterns or using traditional machine learning methods (Logistic Regression, Gaussian-NB, Random Forest, etc.) does not work well. On the other hand, the graph structure can better adapt to intranet data, thus graph-based insider threat detection methods have become mainstream. In order to study the design and effectiveness of graph-based insider threat detection, in this paper, we conduct a systematic and comprehensive survey of existing related research. Specifically, we provide a framework and a taxonomy based on the detection process, classifying existing work from three aspects: data collection, graph construction, and graph anomaly detection. We conduct a quantitative analysis of existing representative graph methods and find that the models with more information have better performance. In particular, we discuss the scalability of existing methods to large-scale networks and their feasibility in real environments. Based on the survey results, we propose 7 pain points in this field and provide specific future research directions. Our survey will provide future researchers with a complete solution.

Multi-view clustering via double spaces structure learning and adaptive multiple projection regression learning

Multi-view clustering aims to group objects with high similarity into one group according to the heterogeneous features of different views. The graph-based clustering methods have obtained excellent results. However, there remain a few common drawbacks. For example, some methods do not consider graphs' high-order structure information. Thus, fuller data information cannot be obtained. In addition, some methods remove noise, outliers, and redundant information in the graph learning phase, resulting in the loss of graph information. Furthermore, using predefined graphs cannot exploit complementary information between views. A triple strategy-based multi-view clustering method is presented to solve the above issues. First, Laplacian graphs are used for fusion learning, and the underlying first-order and second-order structure information among views are explored simultaneously. Then, a label fusion scheme is designed to eliminate noise, outliers, and redundant information and to mine the intrinsic characteristics of data labels. Besides, the consistent label matrix in adaptive regression learning is used to explore complementary information between views in a mutually guided learning way. Finally, the objective function is solved by using an efficient iterative method. Six types of experiments are conducted on eleven real-world multi-view datasets, and the conclusions that can be drawn are: (1) the proposed algorithm achieves the best results in terms of clustering accuracy on ten datasets with an average accuracy improvement of 5.11% compared to other algorithms. Specifically, the accuracy improved by 9.05% on dataset HW and 10.95% on dataset Reuters compared to the second results; (2) The ablation experiments confirm that the different learning strategies included in the proposed algorithm allow it to achieve better clustering performance.

Graph Transformer for 3D point clouds classification and semantic segmentation

Recently, graph-based and Transformer-based deep learning have demonstrated excellent performances on various point cloud tasks. Most of the existing graph-based methods rely on static graph, which take a fixed input to establish graph relations. Moreover, many graph-based methods apply maximizing and averaging to aggregate neighboring features, so that only a single neighboring point affects the feature of centroid or different neighboring points own the same influence on the centroid’s feature, which ignoring the correlation and difference between points. Most Transformer-based approaches extract point cloud features based on global attention and lack the feature learning on local neighbors. To solve the above issues of graph-based and Transformer-based models, we propose a new feature extraction block named Graph Transformer and construct a 3D point cloud learning network called GTNet to learn features of point clouds on local and global patterns. Graph Transformer integrates the advantages of graph-based and Transformer-based methods, and consists of Local Transformer that use intra-domain cross-attention and Global Transformer that use global self-attention. Finally, we use GTNet for shape classification, part segmentation and semantic segmentation tasks in this paper. The experimental results show that our model achieves good learning and prediction ability on most tasks. The source code and pre-trained model of GTNet will be released on https://github.com/NWUzhouwei/GTNet.

Attention-based stackable graph convolutional network for multi-view learning

In multi-view learning, graph-based methods like Graph Convolutional Network (GCN) are extensively researched due to effective graph processing capabilities. However, most GCN-based methods often require complex preliminary operations such as sparsification, which may bring additional computation costs and training difficulties. Additionally, as the number of stacking layers increases in most GCN, over-smoothing problem arises, resulting in ineffective utilization of GCN capabilities. In this paper, we propose an attention-based stackable graph convolutional network that captures consistency across views and combines attention mechanism to exploit the powerful aggregation capability of GCN to effectively mitigate over-smoothing. Specifically, we introduce node self-attention to establish dynamic connections between nodes and generate view-specific representations. To maintain cross-view consistency, a data-driven approach is devised to assign attention weights to views, forming a common representation. Finally, based on residual connectivity, we apply an attention mechanism to the original projection features to generate layer-specific complementarity, which compensates for the information loss during graph convolution. Comprehensive experimental results demonstrate that the proposed method outperforms other state-of-the-art methods in multi-view semi-supervised tasks.